Remote monitoring and fault diagnosis and repair system for offshore wind power construction ship equipment

Abstract

The invention relates to the field of intelligent operation and maintenance of offshore wind power construction ship equipment, and discloses a remote monitoring and fault diagnosis and repair system for offshore wind power construction ship equipment, and the system comprises a data collection module which is used for collecting the operation data of the ship equipment through a multi-mode sensor, and outputting the data to a federal learning module; the federated learning module is used for receiving the output data of the data acquisition module, performing tensor decomposition on the multi-modal data to obtain a local factor matrix, uploading the factor matrix to a cloud server for global aggregation, and issuing the aggregated global factor matrix to the fault diagnosis module; and the resource allocation module is used for dynamically receiving the communication load and the edge computing requirement of the federated learning module. Through a multi-mode sensor network and an anti-interference preprocessing algorithm, the influence of severe environments such as offshore high salt mist and strong electromagnetic interference is effectively overcome, and continuous and stable acquisition of equipment operation data is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of offshore wind power construction ship equipment, and specifically provides a remote monitoring and fault diagnosis and repair system for offshore wind power construction ship equipment. Background Art

[0002] As a key equipment for the construction of offshore wind farms, offshore wind power construction ships are facing complex working conditions such as high salt spray corrosion, extreme sea conditions, and high-load operation of equipment for a long time. The real-time monitoring of the health status of their equipment and the rapid repair of faults are directly related to construction safety and efficiency. Traditional monitoring systems rely on local data collection and manual inspections, with inherent defects such as lagging response, coverage blind spots, and high maintenance costs, and are difficult to meet the remote intelligent operation and maintenance requirements in deep-sea scenarios.

[0003] Current mainstream remote monitoring solutions are mostly based on single-sensor data streams and centralized cloud computing architectures. Although they can realize the cloud analysis of the status of some equipment, significant deficiencies are exposed in practical applications. On the one hand, the harsh offshore environment causes sensor signals to be easily interfered by high-frequency noise, with poor data transmission stability, and it is difficult to synchronize and fuse multi-modal heterogeneous data in time, making it difficult to build a high-precision digital twin model of the equipment. On the other hand, centralized data processing has the risk of privacy leakage, and the allocation of bandwidth and computing power resources is rigid, unable to adapt to dynamic communication loads and edge computing requirements. In addition, traditional fault diagnosis models rely too much on data-driven, lack the embedding of equipment physical laws, resulting in a relatively high false alarm rate, and emergency response mechanisms mostly rely on preset rule libraries, lacking the ability of autonomous decision-making in case of communication interruption, further restricting the reliability of the system.

[0004] How to achieve the efficient collection and anti-interference processing of multi-source data in complex offshore environments and establish a distributed intelligent analysis framework that takes into account privacy protection and collaborative modeling has become the core challenge to improve the real-time performance and reliability of remote monitoring. The data synchronization error, communication resource competition, and insufficient model generalization ability of the existing technology lead to delays in equipment status perception and diagnostic deviations, seriously restricting the fault response efficiency and operation and maintenance safety of construction ships. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a remote monitoring and fault diagnosis and repair system for offshore wind power construction ship equipment, which solves the problems of poor real-time performance of equipment status monitoring and high false alarm rate caused by unstable multi-source data collection and difficult heterogeneous information fusion in complex working conditions such as high salt spray and strong interference in the existing technology.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A remote monitoring and fault diagnosis and repair system for offshore wind power construction ship equipment, comprising: A data acquisition module, configured to collect the operation data of marine equipment through multimodal sensors and output the data to the federated learning module; A federated learning module, which receives the output data of the data acquisition module, performs tensor decomposition on the multimodal data to obtain a local factor matrix, and uploads the factor matrix to a cloud server for global aggregation. The aggregated global factor matrix is then sent to the fault diagnosis module; A resource allocation module, which dynamically receives the communication load and edge computing requirements of the federated learning module, generates bandwidth and computing power allocation instructions, and feeds them back to the federated learning module and the data acquisition module to optimize data transmission; A fault diagnosis module, which receives the global factor matrix of the federated learning module, outputs the fault type and probability through a hybrid model embedding the physical equations of the equipment, and transmits the diagnosis result to the repair decision module; A repair decision module, which generates a repair strategy by invoking the associated physical equation in the digital twin according to the fault type of the fault diagnosis module, selects the optimal strategy through the Monte Carlo tree search algorithm, and sends it to the marine equipment for execution; An emergency response module, which monitors the communication link status. If the communication interruption exceeds a predetermined threshold, it takes over the control of the repair decision module and generates emergency repair instructions based on the reinforcement learning model pre-trained at the edge.

[0007] Preferably, the data acquisition module includes: Deploy anti-salt fog sensors at preset positions of the gearbox, hydraulic pump, winch, pitch system, and generator set of the marine equipment to collect the vibration acceleration , temperature , pressure , current intensity , rotational speed of the time series data in real time; Perform anti-interference preprocessing on the sensor data: Eliminate high-frequency noise through moving average filtering. The filtering formula is: ; where is the original signal, is the filtered signal, and is the size of the moving window; Fill the missing data segment with linear interpolation. The interpolation formula is: ; where , are the valid data points before and after the missing time point respectively.

[0008] Preferably, the data acquisition module further includes: Divide the preprocessed multi-modal data according to a time window to construct a third-order tensor , where: Time step When it reaches a predetermined threshold, it corresponds to 100 sampling points per second; Sensor type dimension When it reaches a predetermined threshold, it includes vibration, temperature, pressure, current, and rotational speed; Device location dimension When it reaches a predetermined threshold, it corresponds to a gearbox, a hydraulic pump, a hoisting winch, a pitch system, and a generator set; Synchronize the heterogeneous sampling rate data. If the sampling rate of a certain sensor reaches a predetermined threshold, then resample its data to a unified time step through cubic spline interpolation. The interpolation function satisfies: ; where, is the original sampling time point, is the corresponding sampling value.

[0009] Preferably, the federated learning module includes: Perform CP decomposition on the local tensor of each device to obtain factor matrices: ; where, , , are factor vectors of the time, sensor type, and device location modalities respectively, is the tensor rank, which is iteratively optimized to the convergence condition .

[0010] Preferably, the federated learning module further includes: Each device only uploads the factor matrix to the central server, shielding the original data , to achieve privacy protection; The server allocates aggregation weights according to the data volume of each device , and calculates the global factor matrix: , , ; The global factor matrix , , is sent to each device, and each device updates the local model through tensor reconstruction .

[0011] Preferably, the resource allocation module includes: Construct a multi-objective optimization function: ; The constraint conditions are: ; Among them, are respectively the data volume, bandwidth, computing power, tensor rank, and reconstructed tensor of the th device; Discretize and into binary variables: ; Construct and solve the Hamiltonian of the QUBO model and solve: ; Decode the quantum annealing result to obtain the optimal bandwidth and computing power , and send them to the federated learning module and the data acquisition module.

[0012] Preferably, the fault diagnosis module includes: Construct a hybrid loss function: ; Among them, is the third-order tensor input by the data acquisition module; is the tensor reconstructed by the federated learning module; is the th time modal factor vector; is the discretized form of the device physical equation, defined by at least one of the gearbox vibration equation and the hydraulic pump pressure-flow equation; is the physical residual weight coefficient; similarly: Minimize by the gradient descent method, and iteratively update the hybrid model parameters until convergence; Take the output of the converged model as the fault type probability distribution, where represents the probability of the th type of fault; When there is , it is determined as the th type of fault and trigger the repair decision module.

[0013] Preferably, the repair decision module includes: According to the fault probability output by the fault diagnosis module , calculate the dynamic repair priority score: ; Among them, is the probability of the type of fault; is the preset fault severity weight; is the historical average repair time; is the smoothing factor; Based on the priority score , construct an integer programming model for resource allocation: ; Among them, indicates whether to repair the type of fault; is the number of robots and spare parts required to repair the type of fault; is the total available robot and spare part resources; Use the branch and bound method to solve the model and generate the optimal repair sequence ; According to , drive the maintenance robot to perform the repair action and update the device status to the federated learning module.

[0014] Preferably, the emergency response module includes: Real-time monitor the communication link status. If the heartbeat signal of the federated learning module is not received for more than a predetermined threshold continuously, it is determined that the communication is interrupted; Call the deep Q-network model pre-trained at the edge, input the current device status tensor and the historical fault records, and output the emergency repair instruction When it exceeds the predetermined threshold, it means to immediately execute the preset repair procedure for the type of fault; Send to the local controller of the ship equipment for execution, and cache the execution result in the edge database and synchronize it to the cloud after the communication is restored.

[0015] Preferably, the system further includes a performance optimization module: Statistical model reconstruction error of the federated learning module and false alarm rate of the fault diagnosis module , construct an adaptive learning rate update rule: ; Among them, is the preset maximum allowable false alarm rate; dynamically adjust the learning rate of the gradient descent method and feed the updated learning rate back to the fault diagnosis module.

[0016] The present invention provides a remote monitoring, fault diagnosis and repair system for offshore wind power construction ship equipment. It has the following beneficial effects: 1. Through the multi-modal sensor network and anti-interference preprocessing algorithm, the present invention effectively overcomes the adverse environmental impacts such as high salt fog and strong electromagnetic interference at sea, ensuring the continuous and stable acquisition of equipment operation data. Combining the tensor decomposition and global aggregation mechanism under the federated learning framework, it realizes the deep fusion and feature extraction of multi-source heterogeneous data, significantly enhancing the system's perception ability for complex working conditions.

[0017] 2. Through the local tensor decomposition and factor matrix uploading mechanism based on federated learning, the present invention avoids the original data leaving the device local, fundamentally solving the risk of sensitive data leakage. Through the weighted aggregation and distribution of the global factor matrix, it realizes multi-device collaborative modeling while protecting privacy, taking into account data utilization and security compliance.

[0018] 3. The present invention adopts a multi-objective optimization model driven by quantum annealing to dynamically allocate communication bandwidth and edge computing power resources, accurately balancing the data transmission delay and computing accuracy requirements. Combining the repair priority evaluation mechanism under resource constraints, it maximizes the utilization efficiency of limited maintenance resources and shortens the fault response cycle.

[0019] 4. Through the hybrid loss function that fuses the data-driven model and the device physical equation, the present invention embeds the device operation law into the diagnostic algorithm, reducing the false alarm rate and missed alarm rate. Based on the dynamic probabilistic output and physical residual analysis, it provides a quantitative basis for the fault type, supporting the scientific and transparent maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the main framework diagram of the present invention; Figure 2 is one of the system flow schematic diagrams of the present invention; Figure 3 is the other system flow schematic diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Please refer to the attached Figure 1 - attached Figure 3, an embodiment of the present invention provides a remote monitoring and fault diagnosis and repair system for offshore wind power construction ship equipment, including: A data acquisition module, configured to collect operation data of ship equipment through multimodal sensors and output the data to the federated learning module; A federated learning module, which receives the output data of the data acquisition module, performs tensor decomposition on the multimodal data to obtain a local factor matrix, and uploads the factor matrix to the cloud server for global aggregation. The aggregated global factor matrix is sent to the fault diagnosis module; A resource allocation module, which dynamically receives the communication load and edge computing requirements of the federated learning module, generates bandwidth and computing power allocation instructions, and feeds them back to the federated learning module and the data acquisition module to optimize data transmission; A fault diagnosis module, which receives the global factor matrix of the federated learning module, outputs the fault type and probability through a hybrid model embedding the physical equation of the device, and transmits the diagnosis result to the repair decision module; A repair decision module, according to the fault type of the fault diagnosis module, calls the associated physical equation in the digital twin to generate a repair strategy, selects the optimal strategy through the Monte Carlo tree search algorithm and sends it to the ship equipment for execution; An emergency response module, which monitors the communication link status. If the communication interruption exceeds a predetermined threshold, it takes over the control right of the repair decision module and generates emergency repair instructions based on the reinforcement learning model pre-trained at the edge.

[0023] The data acquisition module includes: Deploy anti-salt fog sensors at preset positions of the gearbox, hydraulic pump, hoisting winch, pitch system, and generator set of the ship equipment to collect vibration acceleration in real time , temperature , pressure , current intensity , rotational speed of the time series data; Perform anti-interference preprocessing on the sensor data: Eliminate high-frequency noise through moving average filtering. The filtering formula is: ; Among them, is the original signal, is the filtered signal, is the moving window size; Fill the missing data segment with linear interpolation. The interpolation formula is: ; Among them, , are the valid data points before and after the missing time point respectively.

[0024] The data acquisition module further includes: Dividing the preprocessed multi-modal data by time window to construct a third-order tensor , where: Time step When it reaches the predetermined threshold, it corresponds to 100 sampling points per second; Sensor type dimension When it reaches the predetermined threshold, it includes vibration, temperature, pressure, current, and rotational speed; Device location dimension When it reaches the predetermined threshold, it corresponds to the gearbox, hydraulic pump, hoisting winch, pitch system, and generator set; Synchronize the data with heterogeneous sampling rates. If the sampling rate of a certain sensor reaches the predetermined threshold, then resample its data to a unified time step through cubic spline interpolation, and the interpolation function satisfies: ; where, is the original sampling time point, is the corresponding sampling value.

[0025] The federated learning module includes: Perform CP decomposition on the local tensor of each device to obtain the factor matrices: ; where, , , are the factor vectors of the time, sensor type, and device location modalities respectively, is the tensor rank, which is iteratively optimized to the convergence condition .

[0026] The federated learning module also includes: Each device only uploads the factor matrix to the central server, shielding the original data , to achieve privacy protection; The server allocates aggregation weights according to the data volume of each device , and calculates the global factor matrix: , , ; Send the global factor matrix , , down to each device, and each device reconstructs the tensor through Update the local model.

[0027] The resource allocation module includes: Construct a multi-objective optimization function: ; The constraint conditions are: ; Among them, are respectively the data volume, bandwidth, computing power, tensor rank, and reconstructed tensor of the th device; Discretize and into binary variables: ; Construct and solve the Hamiltonian of the QUBO model and solve: ; Decode the quantum annealing result to obtain the optimal bandwidth and computing power , and send them to the federated learning module and the data collection module.

[0028] The fault diagnosis module includes: Construct a hybrid loss function: ; Among them, is the third-order tensor input by the data collection module; is the tensor reconstructed by the federated learning module; is the th time modal factor vector; is the discretized form of the device physical equation, defined by at least one of the gearbox vibration equation and the hydraulic pump pressure-flow equation; is the physical residual weight coefficient; similarly: Minimize by the gradient descent method, and iteratively update the hybrid model parameters until convergence; Take the output of the converged model as the fault type probability distribution, where represents the probability of the th type of fault; When there is , it is determined as the th type of fault and the repair decision module is triggered.

[0029] The repair decision module includes: According to the failure probability output by the failure diagnosis module , calculate the dynamic repair priority score: ; Among them, is the probability of the th type of failure; is the preset failure severity weight; is the historical average repair time; is the smoothing factor; Based on the priority score , construct an integer programming model for resource allocation: ; Among them, indicates whether to repair the th type of failure; is the number of robots and spare parts required to repair the th type of failure; is the total available robot and spare part resources; Use the branch and bound method to solve the model and generate the optimal repair sequence ; According to , drive the maintenance robot to perform the repair action and update the device status to the federated learning module.

[0030] The emergency response module includes: Real-time monitor the communication link status. If the heartbeat signal of the federated learning module is not received continuously for more than a predetermined threshold, it is determined that the communication is interrupted; Call the deep Q-network model pre-trained at the edge, input the current device status tensor and the historical failure records, and output the emergency repair instruction When it exceeds the predetermined threshold, it means to immediately execute the preset repair procedure for the th type of failure; Send to the local controller of the ship equipment for execution, and cache the execution result in the edge database and synchronize it to the cloud after the communication is restored.

[0031] The system also includes a performance optimization module: Statistical model reconstruction error of the federated learning module and the false alarm rate of the failure diagnosis module , construct an adaptive learning rate update rule: ; Among them, is the preset maximum allowable false alarm rate; dynamically adjust the learning rate of the gradient descent method , and feed back the updated learning rate to the fault diagnosis module.

[0032] In this embodiment, a multi-modal sensor cluster with anti-salt spray corrosion characteristics is deployed at key monitoring positions of the gearbox, hydraulic pump, winch, pitch system and generator set of the ship equipment. Preferably, the sensor types include vibration acceleration sensors, temperature sensors, pressure sensors, current sensors and rotational speed sensors, covering the multi-dimensional physical quantity monitoring of the equipment operating state. Each sensor adopts a sealed packaging structure, filled with inert gas inside and coated with a polytetrafluoroethylene protective layer outside to resist the erosion of the high-salt spray environment at sea. The sensor power supply line and signal transmission line are connected through armored shielded cables to avoid the influence of electromagnetic interference on the data acquisition accuracy.

[0033] Aiming at the problem that sensor signals are vulnerable to high-frequency noise interference under complex offshore working conditions, this embodiment preprocesses the original data by combining moving average filtering and linear interpolation. Specifically, for the vibration acceleration signal, the time series data is smoothed through a moving window, and the filtering formula is expressed as: ; where is the original signal, is the filtered signal, is the moving window size. Preferably, the window size is dynamically adjusted according to the main frequency range of the equipment vibration signal to adapt to the characteristic frequencies of different equipment. For local data missing caused by instantaneous sensor failure or communication packet loss, a linear interpolation algorithm is used for filling. The interpolation formula is defined as: ; where and are the adjacent valid data points before and after the missing time point respectively. Through this interpolation method, the data continuity can be effectively restored, avoiding deviation in subsequent analysis due to data missing.

[0034] Due to the differences in the sampling rates of different sensors (for example, vibration sensors usually use high-frequency sampling while temperature sensors have a lower sampling rate), this embodiment realizes the time series alignment of multi-modal data through a cubic spline interpolation algorithm. Specifically, for sensor data with a sampling rate lower than the reference frequency , an interpolation function that satisfies the second derivative continuity is constructed, and its mathematical expression is: ; where is the original sampling time point, Is the corresponding sampled value. Through this interpolation method, the low-frequency data is resampled to a unified time base to ensure strict synchronization of multimodal data in the time dimension.

[0035] Furthermore, the synchronized multimodal data is divided into a third-order tensor structure according to a fixed time window. The dimensions of the tensor are defined as time steps, sensor types, and device locations, and the mathematical representation is: ; Among them, the time dimension Corresponds to the number of time series points collected per second, and the sensor type dimension Covers five physical quantities: vibration, temperature, pressure, current, and rotational speed. The device location dimension Is mapped to five key devices: gearbox, hydraulic pump, hoisting winch, pitch system, and generator set. Through the tensor structure, this embodiment realizes the unified representation of multi-source heterogeneous data and provides a structured input for subsequent federated learning and fault diagnosis.

[0036] In this embodiment, the federated learning module receives the third-order tensor output by the data acquisition module , where the time dimension Corresponds to the fixed time window divided by the data acquisition module, and the sensor type dimension Covers five physical quantities: vibration, temperature, pressure, current, and rotational speed. The device location dimension Is mapped to five monitoring locations: gearbox, hydraulic pump, hoisting winch, pitch system, and generator set.

[0037] To achieve a low-rank representation of high-dimensional data, perform CP decomposition (Canonical Polyadic Decomposition) on the local tensor Of each device, and the mathematical expression is: ; Among them, , , Are the factor vectors of the time modality, sensor type modality, and device location modality respectively, Is the tensor rank parameter used to control the model complexity. Preferably, optimize each factor matrix by the Alternating Least Squares (ALS) iteration, fix the factor matrices of two modalities in turn and update the factor vectors of the third modality until the convergence condition is satisfied: ; Among them, Is the reconstructed tensor, Is the preset reconstruction error threshold, and its value is associated with the norm of the original tensor output by the data acquisition module and is defined as 。

[0038] In this embodiment, each device only uploads the factor matrices 、 、 locally decomposed to the cloud server, and the original tensor is retained locally to ensure data privacy. The cloud server calculates the aggregation weights according to the data volume of each device, and generates the global factor matrix by weighted average according to the weights: , , ; After the global factor matrix 、 、 is sent to each device, the local model is updated through tensor reconstruction: ; This reconstruction process combines the global features with the local data distribution to achieve collaborative modeling under the federated learning framework.

[0039] In this embodiment, the federated learning module receives the bandwidth and computing power allocation instructions sent by the resource allocation module, and dynamically adjusts the factor matrix upload frequency and decomposition iteration times. Preferably, when the bandwidth resource is limited, the top factor vectors that contribute the most to the reconstruction error are preferentially uploaded to reduce the communication overhead. The computing power resource allocation achieves dynamic load balancing by limiting the maximum iteration times of the ALS algorithm.

[0040] In this embodiment, the resource allocation module receives the communication load parameters (including data volume , tensor rank , and reconstruction error ) uploaded by the federated learning module of each device, and constructs an objective function for jointly optimizing communication delay and data quality loss: ; wherein, is the bandwidth allocation amount (unit: Mbps) of the th device, is the computing power allocation amount (unit: GFLOPS), is the weight coefficient, which is used to balance the relative importance of communication delay and data quality loss. Preferably, the weight coefficient is dynamically adjusted according to the reconstruction error threshold of the federated learning module, and increases when the reconstruction error exceeds the preset threshold To prioritize ensuring data quality. The constraint conditions include upper limit constraints on total bandwidth and total computing power: ; And the lower limit constraint of single-device resources: ; QUBO model conversion and discretization processing: To adapt to the quantum annealing solver, in this embodiment, continuous variables and are discretized into binary variables. Specifically, the bandwidth allocation is discretized in units of 1Mbps: ; The computing power allocation is discretized in units of 10GFLOPS: ; Among them, , , and the discretization granularity is dynamically determined according to the total amount of resources.

[0041] Mapping the objective function and constraint conditions to the Hamiltonian of the QUBO model: ; Among them, the linear term coefficient and the quadratic term coefficient are converted from the communication delay and data quality loss terms in the objective function. The penalty coefficients and are determined by the Lagrange multiplier method to ensure strict satisfaction of the constraint conditions. Preferably, a D-Wave quantum annealing machine is used to solve the Hamiltonian to obtain the optimal binary variable combination , and after decoding, the optimal allocation scheme of bandwidth and computing power is obtained.

[0042] In this embodiment, the resource allocation module sends the optimal allocation scheme and to the federated learning module to dynamically adjust its communication and computing strategies: Communication optimization under bandwidth limitation: When is lower than the default upload bandwidth of the federated learning module, preferentially upload the top factor vectors that contribute the most to the reconstruction error to reduce the amount of communication data; Computing load balancing under computing power allocation: According to Adjust the maximum number of iterations of the Alternating Least Squares (ALS). When computing power is sufficient, increase the number of iterations to improve the decomposition accuracy. When computing power is limited, reduce the number of iterations to reduce the calculation latency.

[0043] In this embodiment, the fault diagnosis module receives the global factor matrix issued by the federated learning module and constructs the original tensor based on the data acquisition module and the reconstructed tensor to construct a hybrid loss function: ; where the first term is the data reconstruction error, which measures the reconstruction accuracy of the federated learning model; the second term is the physical residual constraint, which ensures that the time evolution law of the factor vector conforms to the actual physical law by embedding the device physical equation . The physical equation is dynamically selected according to the device type: the vibration equation of the gearbox: ; where is the damping coefficient, is the stiffness coefficient, which is determined by the device design parameters; ; where is the flow coefficient, which is related to the structural parameters of the hydraulic pump. Preferably, the physical residual weight coefficient is determined by cross-validation to balance the contribution ratio of data-driven and physical constraints. Gradient descent optimization and fault probability output: Minimize the hybrid loss function by the gradient descent method, and iteratively update the model parameters until convergence. Specifically, the Adam optimizer is used to dynamically adjust the learning rate, and the initial learning rate is set to , and it is adaptively adjusted according to the change rate of the loss function after each iteration. Preferably, when the loss decrease rate of 10 consecutive iterations is lower than , it is determined to be convergent. After the model converges, output the fault probability vector , where represents the occurrence probability of the th type of fault. The probability calculation is based on normalizing the joint distribution of the reconstruction error and the physical residual by the Softmax function: ; where is the typical reconstruction error threshold corresponding to the th type of fault, is the corresponding physical residual threshold, which is obtained by statistical analysis of historical fault data. When there is , it is determined to be the th type of fault and trigger the repair decision module.

[0044] In this embodiment, the repair decision-making module receives the fault probability vector output by the fault diagnosis module , combines the preset fault severity weight and the historical repair time data, and calculates the dynamic repair priority score: ; Wherein, is the probability of the th type of fault, which is output by the fault diagnosis module based on the hybrid loss function; is the preset fault severity weight, and the weight value is manually set according to the equipment downtime risk and safety level; is the historical average repair time (unit: hour), which is obtained by statistically analyzing past maintenance records; is the smoothing factor, which is used to avoid a zero denominator. Preferably, , to balance the priority calculation of new fault types.

[0045] Based on the priority score , an integer programming model that maximizes the total repair priority is constructed: ; Wherein, is a binary decision variable, indicating whether to repair the th type of fault; and are respectively the number of robots and the number of spare parts required to repair the th type of fault, which are predefined by the equipment maintenance manual or expert experience; and are the total number of available robots and the number of spare parts currently, which are updated in real time by the resource allocation module.

[0046] Branch and bound method for solving and repair execution: The branch and bound algorithm is used to solve the integer programming model to generate the optimal repair sequence . Specifically, the feasible solution space is recursively divided into subsets (branching), combined with the upper and lower bound estimation of the relaxation problem (bounding), and non-optimal solutions are gradually excluded. Preferably, the fault branches with high values are processed first to accelerate convergence.

[0047] After the solution is obtained, according to drive the maintenance robot to execute the repair action: If , call the Class fault repair program, control the robot to perform spare part replacement, component calibration or lubrication maintenance according to the preset path; after the repair is completed, update the device status to the federated learning module, and trigger the data acquisition module to re-acquire the operation data to verify the repair effect.

[0048] When the communication interruption triggers the emergency response module, the repair decision module switches to the local cache mode, and based on the historical optimal strategy stored at the edge Execute the repair, and synchronize with the cloud strategy and calibrate the priority score after the communication is restored.

[0049] In this embodiment, the emergency response module continuously monitors the heartbeat signal with the cloud federated learning module. If the duration of continuously not receiving the heartbeat signal exceeds the preset threshold (preferably, the threshold is 60 seconds), it is determined as a communication interruption. After the interruption is triggered, immediately take over the control of the repair decision module and switch to the edge autonomous decision-making mode.

[0050] In this embodiment, the pre-trained deep Q-network (DQN) model is deployed on the edge computing node, and the input is the current device status tensor: And historical fault records (where is the number of historical faults, is the total number of fault types), Output the emergency repair instruction vector . The DQN model updates the strategy through the Q-value function: ; where the state is and The joint feature encoding of, the action corresponds to The component of, the reward function is calculated according to the similarity between the device status after repair and the historical normal status.

[0051] Emergency instruction execution and result caching: When , call the preset repair program of the th type of fault stored locally, and drive the maintenance robot to perform standardized operations (including emergency shutdown, standby system switching, redundant component activation, etc.). The sensor data and operation logs during the execution are cached in real time to the edge database, and the data format is compatible with the cloud federated learning module. After the communication is restored, automatically synchronize the cached data to the cloud, and trigger the global model update and priority score calibration.

[0052] In this embodiment, the emergency response module and the repair decision module share resource constraint parameters (including and ), dynamically adjust the repair strategy based on the local resource pool during communication interruption. Preferably, a Receding Horizon Control (RHC) mechanism is adopted to re-evaluate the available resources and fault status every 5 minutes and update the emergency instructions to adapt to real-time changes.

[0053] In this embodiment, the communication interruption detection ensures the real-time perception of the system state, the DON model realizes intelligent decision-making at the edge, and the emergency instruction execution and caching mechanism ensure the continuity and traceability of the repair operation. The above technical features form a closed loop to maintain the safe operation of the device under extreme conditions without cloud support.

[0054] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A remote monitoring, fault diagnosis and repair system for offshore wind power construction ship equipment, characterized in that Including: A data acquisition module, which is used to collect the operation data of ship equipment through multimodal sensors and output the data to the federated learning module; A federated learning module, which receives the output data of the data acquisition module, performs tensor decomposition on the multimodal data to obtain a local factor matrix, and uploads the factor matrix to the cloud server for global aggregation. The aggregated global factor matrix is sent to the fault diagnosis module; A resource allocation module, which dynamically receives the communication load and edge computing requirements of the federated learning module, generates bandwidth and computing power allocation instructions, and feeds them back to the federated learning module and the data acquisition module to optimize data transmission; A fault diagnosis module, which receives the global factor matrix of the federated learning module, outputs the fault type and probability through a hybrid model embedded with the physical equation of the device, and transmits the diagnosis result to the repair decision module; A repair decision module, which, according to the fault type of the fault diagnosis module, calls the associated physical equation in the digital twin to generate a repair strategy, selects the optimal strategy through the Monte Carlo tree search algorithm, and sends it to the ship equipment for execution; An emergency response module, which monitors the communication link status. If the communication is interrupted for more than a predetermined threshold, it takes over the control of the repair decision module and generates emergency repair instructions based on the reinforcement learning model pre-trained at the edge; 2. The remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, wherein The data acquisition module includes: Deploy anti-salt spray sensors at preset positions of the gearbox, hydraulic pump, winch, pitch system, and generator set of the ship equipment to collect vibration acceleration in real time , temperature , pressure , current intensity , rotational speed of the time series data; Performing anti-interference preprocessing on the sensor data: Eliminating high-frequency noise through moving average filtering, and the filtering formula is: ； Among them, is the original signal, is the filtered signal, is the sliding window size; Filling the missing data segment with linear interpolation, and the interpolation formula is: ； Among them, , are respectively the valid data points before and after the missing time point .

3. A remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, characterized in that, The data acquisition module further includes: Divide the preprocessed multimodal data according to time windows to construct a third-order tensor , where: Time step When the predetermined threshold value is reached, it corresponds to 100 sampling points per second; Sensor type dimension When the predetermined threshold value is reached, it includes vibration, temperature, pressure, current, and rotational speed; Device location dimension When the predetermined threshold value is reached, it corresponds to the gearbox, hydraulic pump, hoisting winch, pitch system, and generator set; Synchronize heterogeneous sampling rate data. If the sampling rate of a certain sensor is lower than a predetermined threshold, resample its data to a unified time step through cubic spline interpolation. The interpolation function satisfies: ； Among them, is the original sampling time point, is the corresponding sampling value.

4. An offshore wind power construction vessel equipment remote monitoring and fault diagnosis and repair system according to claim 1, characterized in that, The federated learning module includes: Local tensors for each device Perform CP decomposition to obtain factor matrices: ； Among them, , , are the factor vectors of time, sensor type, and device location mode respectively, is the tensor rank, which is iteratively optimized to the convergence condition by the alternating least squares method .

5. A remote monitoring, fault diagnosis and repair system for offshore wind power construction ship equipment according to claim 1, characterized in that, The federated learning module also includes: Each device only uploads the factor matrix to the central server, masking the original data , achieving privacy protection; The server assigns aggregation weights based on the data volume of each device and calculates the global factor matrix: , , , ； Send the global factor matrix , , to each device, and each device updates the local model through tensor reconstruction .

6. The remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, wherein, The resource allocation module includes: Constructing a multi-objective optimization function: ； The constraint condition is: ； Among them, are respectively the data volume, bandwidth, computing power, tensor rank, and reconstructed tensor of the th device; Discretize and into binary variables: ； Construct and solve the Hamiltonian of the QUBO model And solve for: ； Decoding the results of quantum annealing to obtain the optimal bandwidth and computing power , and send them to the federated learning module and the data acquisition module.

7. A remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, characterized in that, The fault diagnosis module includes: Constructing a hybrid loss function: ； Among them, is a third-order tensor input by the data acquisition module; Tensors for the reconstruction of the federated learning module; is the th time-modal factor vector; It is the discretized form of the physical equation of the device, defined by at least one of the vibration equation of the gearbox and the pressure-flow equation of the hydraulic pump; is the physical residual weight coefficient; similarly: Minimize by gradient descent method and iteratively update the hybrid model parameters until convergence; The output of the converged model is used as the probability distribution of the fault type, where represents the probability of the -th type of fault; When there is , it is determined as the type of fault and the repair decision module is triggered.

8. A remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, characterized in that The repair decision module includes: According to the failure probability output by the failure diagnosis module , calculate the dynamic repair priority score: ； Among them, is the probability of the type of fault; is the preset fault severity weight; is the historical average repair time; is the smoothing factor; Based on the priority score , a resource allocation integer programming model is constructed: ； Among them, indicates whether to repair the type of fault; is the number of robots and the number of spare parts required to repair the type of fault; is the total available robot and spare part resources; Use the branch and bound method to solve the model and generate the optimal repair sequence ; According to Drive the maintenance robot to perform repair actions and update the device status to the federated learning module.

9. The remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, characterized in that, The emergency response module includes: Real-time monitoring of the communication link status. If the heartbeat signal of the federated learning module is not received continuously for more than a predetermined threshold, it is determined that the communication is interrupted; Call the deep Q-network model pre-trained at the edge, and input the current device state tensor and the historical fault records, and output emergency repair instructions When it exceeds the predetermined threshold, it means to immediately execute the preset repair procedure for the nth type of fault; Send to the local controller of the ship equipment for execution, and cache the execution result in the edge database, and synchronize it to the cloud after the communication is restored.

10. The remote monitoring, fault diagnosis and repair system for an offshore wind power construction ship equipment according to claim 1, wherein The system further includes a performance optimization module: Statistically analyze the model reconstruction error of the federated learning module and the false alarm rate of the fault diagnosis module , and construct an adaptive learning rate update rule: ； Among them, is the preset maximum allowable false alarm rate; dynamically adjust the learning rate of the gradient descent method , and feedback the updated learning rate to the fault diagnosis module.

Images